← News
AI tutor agents, omnimodal video models, LTX-2 updates, long-term memory, video faceswap: AI NEWS
This episode is a “demo reel” style roundup of new papers, repos, and model updates. The quickest way to use it is to pick one item below and try to reproduce the author’s core demo.
- DreamID‑V: identity-driven video editing / face swap style workflows.
- UniVideo: a unified video generation/editing approach (worth skimming for the training/data choices).
- SimpleMem: long-term memory for agents, with emphasis on retrieval and “what to remember” policies.
- DreamStyle: style transfer / identity styling across images (pairs nicely with the identity/video work).
- DeepTutor: tutoring-agent scaffolding from HKU; useful if you’re building “explain, quiz, correct” loops.
- NeoVerse: 4D / dynamic scene work; interesting for anyone tracking “video → world model” progress.
- LTX‑2 updates (plus GGUF distribution): practical note if you’re running local video models.
- HY‑MT: Tencent’s multimodal translation model; a reminder that strong MT is still a core building block for multimodal apps.